1. Share the Reel
Send or forward a cooking Reel to the CraveLab bot on Instagram. Nothing is copied by hand: the message, the caption, and any links that came with it are all taken as part of the source.
CraveLab turns cooking Reels into recipes you can cook from, without asking you to trust a model. Extraction is a draft with evidence. You stay the editor.
Send or forward a cooking Reel to the CraveLab bot on Instagram. Nothing is copied by hand: the message, the caption, and any links that came with it are all taken as part of the source.
CraveLab downloads the video, checks it, and stores it in your own private object space alongside its checksum. The recipe always keeps a way back to the clip it came from.
The complete video with audio, the caption, your message, and safely fetched linked text go to Gemini, which returns one structured recipe: ingredients, quantities, ordered steps, timings, temperatures, equipment, storage and reheating notes.
A CraveLab recipe is not a block of text. Each field — a quantity, a step, a resting time — is linked to where it came from: a moment in the video, something said out loud, the caption, your message, or a linked page. Video and audio evidence keeps its start and end timestamp, so you can jump to the second that justifies a number and judge it yourself.
Evidence also records how confident the extraction was. That is what makes the recipe reviewable instead of merely plausible.
When something cannot be read off the Reel, CraveLab says so. Unclear or missing details are recorded as uncertainties and surfaced on the recipe rather than quietly filled in with a guess. If an on-screen quantity or oven temperature is missing, the import can retry with high-resolution frames from the video before giving up; if it still cannot be resolved, the import waits for you in the inbox.
Your corrections never overwrite what the extraction produced. Saving a change writes a new version of the recipe with a short summary of what you changed, and the previous versions stay readable. Provider records are kept as they were received, so the history of a recipe stays honest.
Gemini output never becomes a nutrient value. Every ingredient that contributes to nutrition has to be matched to a real record, and you confirm the match:
Totals are then calculated from the matched records and the scaled quantities, so a portion count you change is reflected honestly instead of re-estimated.
Recipes live in libraries. You can invite someone as an editor, who versions the shared original with you, or as a viewer. When a viewer edits a shared recipe, the change is forked into their own private library instead of altering yours — they get a recipe of their own, and your version is untouched.
Nothing you import is public. Your recipes, imports, pantries and cooking history are readable only by you and the people you deliberately share a library with, enforced in the database itself rather than only in the application. Original media is served through short-lived links that check your access first.
You can export your account from Settings at any time, and erase it from the same place: erasure removes your account, your Instagram link, your private libraries and pantries, and the stored media and derived artifacts. Content owned by someone else stays with them, with your membership removed.
CraveLab is invitation-only while it is in beta. Sign in to claim the recipes the bot has already saved for you, or to link your Instagram account.
Recipe extraction and ingredient matching can be wrong. Check allergies, food safety and cooking temperatures before you rely on them.